Skip to main content

Table 2 Estimated predictive performance of logistic regression models in severe fever with thrombocytopenia syndrome, Miyazaki Prefecture

From: Spatial epidemiological determinants of severe fever with thrombocytopenia syndrome in Miyazaki, Japan: a GWLR modeling study

All subjects (n = 97 mesh)

LR (all cases)

GWLR (all cases)

LR (cases by 2016)

GWLR (cases by 2016)

Proportion (%)

22.7 (14.3, 31.0)

22.7 (14.3, 31.0)

6.2 (1.4, 11.0)

6.2 (1.4, 11.0)

Sensitivity (%)

86.4 (72.0, 100)

90.9 (78.9100)

83.3 (53.5100)

83.3 (53.5, 100)

Specificity (%)

61.3 (50.3, 72.4)

58.7 (47.5,69.8)

75.8 (67.0,84.6)

73.6 (64.6, 82.7)

PPV (%)

39.6 (31.7, 47.5)

39.2 (32.1,46.4)

18.5 (10.8,26.3)

17.2 (10.2, 24.3)

NPV (%)

93.9 (87.7, 100)

95.7 (90.1100)

98.6 (96.0,100)

98.5 (95.9100)

Precision (%)

39.6 (25.7, 53.4)

37.7 (24.7, 50.8)

19.2 (4.1, 34.4)

19.2 (4.1, 34.4)

F1-score

0.54

0.53

0.31

0.31

AUC (%)

72.4 (62.7, 80.3)

73.9 (63.5, 80.9)

75.6 (66.2, 83.1)

76.6 (67.2,83.9)

  1. Abbreviations: LR Logistic regression, GWLR Geographically weighted logistic regression, PPV Positive predictive value, NPV Negative predictive value, F1-score Harmonic average of the precision and recall, AUC Area under the receiver operating characteristic curve
  2. Values in parentheses are 95% confidence intervals
  3. All cases represent prediction using all available datasets at the end of 2017. To avoid overly optimistic results, we also predicted 2017 cases using data at the end of 2016 (Cases by 2016)